Dynamic adaptivity of a computer system is its ability to modify the behavior according to the environment in which it is executed. It allows the system to achieve better performance, but usually requires specialized architecture and brings more complexity. The thesis presents an analysis and design of a framework that allows simple and fluent performance-based adaptive development at the level of functions and methods. It closely examines the API requirements and possibilities of integrating such a framework into the Scala programming language using its advanced syntactical constructs. On theoretical level, it deals with the problem of selecting the most appropriate function to execute with given input based on measurements of previous executions. In the provided framework implementation, the main stress is laid on modularity and extensibility, as many possible future extensions are outlined. The solution is evaluated on a variety of development scenarios, ranging from input adaptation of algorithms to environment adaptations of complex distributed computations in Apache Spark.

Modern High Performance Computing (HPC) systems are made up of thousands of server-grade compute nodes linked through a high-speed network interconnect. Each node has tens or even hundreds of CPU cores each, with counts continuing to grow on newer HPC clusters. This results in a need to make use of millions of cores per cluster. Fully leveraging these resources is difficult. There is an active need to design software that scales and fully utilizes the hardware. In this dissertation, we address this gap with a dual approach, considering both intra-node (single node) and inter-node (across node) concerns. To aid in intra-node performance, we propose two novel concurrent data structures: a transactional vector and a persistent hash map. These designs have broad applicability in any multi-core environment but are particularly useful in HPC, which commonly features many cores per node. For inter-node performance, we propose a metrics-driven approach to improve scheduling quality, using predicted run times to backfill jobs more accurately and aggressively. This is augmented using application input parameters to further improve these run time predictions. Improved scheduling reduces the number of idle nodes in an HPC cluster, maximizing job throughput. We find that our data structures outperform the prior state-of-the-art while offering additional features. Our backfill technique likewise outperforms previous approaches in simulations, and our run time predictions were significantly more accurate than conventional approaches. Code for these works is freely available, and we have plans to deploy these techniques more broadly on real HPC systems in the future.